[
https://issues.apache.org/jira/browse/HADOOP-12066?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14577050#comment-14577050
]
Kai Zheng commented on HADOOP-12066:
------------------------------------
Thanks [~vinayrpet] for the review and comments!
bq. Anyway If needed to move to HDFS project, I am OK with that too.
Yeah, if anyone would help convert this, I'm OK with it.
bq. Is the intention of Jira is just to allow configuring the factory class?
and generate the raw coders from the configured factory classes instead of
directly calling constructors?
You're right. It's basically to configure RS and XOR raw coders, and
configuring the coder factory class is the mean. This is good for both near
term and long term. For now we need this because we're using raw coders
directly in the first phase; For the long term the {{RSErasureCodec}} will also
need this to determine what kinds of raw RS or XOR coder to use (either the
Java one or any native one like in HADOOP-11540). This isn't overlapping with
codec configuration. In future, we will need to configure the erasure codec to
use, and for some certain codec, we will also need to configure what raw
coder(s) to use in the underlying behavior. This is doing something for the
later. Please let know if this doesn't clarify the confusing. Thanks.
Will double check the patch and correct the formats.
> Allow to configure RS and XOR raw coders
> -----------------------------------------
>
> Key: HADOOP-12066
> URL: https://issues.apache.org/jira/browse/HADOOP-12066
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: io
> Reporter: Kai Zheng
> Assignee: Kai Zheng
> Attachments: HADOOP-12066-HDFS-7285-v1.patch
>
>
> This is to allow configuring RS and XOR raw coders as they're the basic ones
> and used in other advanced codecs like HH and LRC, and we're having both
> implementations in Java and native.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
